Table of Contents

Goals of Analysis

Visualize the following

  1. Conversion Rate
  2. Average Basket Size
  3. Cost-to-Income Ratio
  4. Any other relevant findings, insights or comparisons

Relevant Formulas Given

image.png

Initial Exploration

Transform Visits, Orders, Revenue in \$ to Numerical types

Add the 3 columns for metrics

  1. Conversion Rate --> percentage
  2. Average Basket Size --> \$
  3. Cost-to-Income Ratio --> ratio, but let's express as a % - e.g. if cost_to_income ratio is 9%, company has to spend 9 cents to make a dollar

Revenue per Visitor (RPV)

Some quick insights so far from 'mean' row of describe

Week 42 vs Week 43

  1. Mean CIR is lower for 43: 1.23% as opposed to 2.01%: this means in week 43 the company had to spend 78 cents less to make 1$ of revenue from the average channel. This could be the result of particular promos/campaigns in Week 43.

  2. Mean RVP in week 43 is higher: 66 cents per visitor for the avg channel vs 53 cents

  3. Mean Spend is lower in Week 43 - could this be because week 42 spend includes some setup/initial fixed costs for these channels? While this could be the cause of slightly less mean visits per channel, mean of orders per channel goes up

Let's confirm if 'Paid' category leads to better metrics

Group by Week and Paid/Free Category

Plotly Visualizations - Paid vs Free Metrics

All Plotly visualizations are interactive - you can hover over the individual bars or pie chart sections for more information.

As expected CR, ABS and RPV are higher for the paid category channels, in both weeks.

Pitfall of CIR (inverse of Return-on-Ad Spend) when comparing Paid v Free

Though from the viz below, it may appear the 'free' categories have a much better (lower) cost to income ratio, it's important to note that these channels (like Newsletters), may saturate after a point.

Spending on newsletters to the level of paid channels would not bring paid-channel level returns.

Also, 'free' channels like newsletters are meant more to promote brand engagement than increase revenue/conversions. And in fact, spamming the same users with more newsletters could have negative consequences on brand image.

So this is one visualization that could be misleading without understanding the full context.

This visualization is also interactive

Plotting Visits vs Orders

It intuitively makes sense that as visits increase, the number of orders should too. And this is the trend observed in the scatter plot below, with a regression line fit.

This visualization is also interactive.

Linear Trendline Observations

When fitting a trendline (OLS - Ordinary Least Squares) of Orders vs Visits, we see that there are a few that don't exactly fit the regression line or are further away from it: for example, in the 3 points between 400 and 500K visitors, there is one that has much higher # of Orders.

Hovering over it, we see that it is for Channel 'App install network B' in Week 43. It is worth looking into what could have caused this increase:

image-2.png

Visualizing CR, ABS, CIR and RPV for each Channel (agg across 42, 43)

This visualization is also interactive

It's easy to see which channels are better performing from these bar graphs -

Top Performers

Low Performers

Revenue Share, Spend Share

This visualization is also interactive

Drilling down into Paid Channels

Let's now take a closer look at paid channels

This visualization is also interactive

Week 42 of Paid Channels

Week 43 of Paid Channels

This visualization is also interactive

Insights on Comparing Week 42 and Week 43

Other Potential Avenues for Analysis

  1. Inclusion of funnel chart, if there are more stages of marketing funnel available: beyond # of visitors, # of orders, maybe also # of app downloads (one stage before visitor), # of second/third order customers, # who refer others
  2. It may also be useful to see cumulative conversation rate % across days of each week - there might be an inflection point/day at which a higher % of conversions are noted
  3. It would be interesting to see in the weeks after 42, 43, if App install network B and C continue to be the dominant channels or not, and the reasons behind this